honestly, most of the time this lora is completely unnecessary as chroma can handle it just fine on its own, but when all else fails.
work in progress.
v0.1a is the original trained model and v0.1a-small is a resized version with mostly the same results (see preview to compare, images for both are same seed/prompt).
Description
resized version of the original v0.1a with mostly the same results, ever so slightly toned down maybe.
FAQ
Comments (2)
I tried it and it had almost no effect in Z-IMAGE. The weight is 1, but it has no effect.
all the above samples were generated with euler/beta with steps 6 if that helps... this lora can be kind of sensitive to certain samplers. res_2s/simple seems to work pretty well too